In Chapter 1 of the Generative AI Passport, we provide a summary for those who have learned these terms in a fragmented way. Once you finish reading, you will see the relationship between these three terms and the single flow that leads from them to generative AI. We will not use any mathematical formulas.

The three terms are nested

The first thing to grasp is that these three terms are not separate items lined up side-by-side, but rather a relationship where smaller boxes are contained within a larger box.

AI (Artificial Intelligence): A general term for technologies and research aimed at realizing functions such as judgment and inference, similar to humans, using computers.
Machine Learning: One of the methods for realizing AI. It involves discovering patterns from data.
Deep Learning: A type of machine learning method. It uses neural networks with many deep layers.

It is easier to understand if you compare it to a company organization. A company has a “Sales Department,” and within that, there is a “Corporate Sales Section.” A person in the Corporate Sales Section is also a person in the Sales Department and a person in the company. Similarly, deep learning is a type of machine learning, and machine learning is a type of AI.

The exam syllabus consists of 5 chapters, with Chapter 1 being “AI (Artificial Intelligence)” and Chapter 2 being “Generative AI.” If you grasp this nesting in Chapter 1, the content of Chapter 2 will be much easier to understand.

Two ways to give AI intelligence

There are broadly two ways to make a computer behave intelligently.

Rule-based

This is a method where humans pre-write rules such as “if this, then do that.” Think of an operations manual given to a new employee. If it says, “Expenses exceeding a certain amount require approval from a supervisor,” they can act accordingly.

Pros: It is easy to explain why a certain judgment was made.
Cons: It cannot handle situations not written in the manual. As exceptions increase, it becomes impossible to write down all the rules.

Machine Learning

Instead of humans writing the rules, this is a method where the computer itself discovers patterns from vast amounts of data. This is closer to the “experience” of a veteran employee who has handled many cases. Even if it is not written in a manual, they can guess that “this inquiry is likely to take a long time” based on similar past examples.

On the other hand, because the judgment is acquired from experience, it is difficult to explain the reason in words. Also, how smart it becomes depends heavily on the quantity and quality of the data it learns from. Think of rule-based and machine learning not as one being superior to the other, but as two methods that excel in different situations.

Three methods of machine learning

Machine learning is divided into three types based on how data is provided.

Supervised Learning

A method of learning using data with correct answers attached. For example, if you label past emails as “spam” or “normal email” and have the system learn from them, it will be able to sort newly arrived emails. This is similar to a state where a senior colleague corrects your work one by one.

Unsupervised Learning

A method of finding patterns or structures from data without correct answers. It is like the task of bundling a pile of business cards on a desk into “similar industries” without being instructed by anyone. It is used for purposes such as dividing customers into several groups.

Reinforcement Learning

A method of learning actions that yield good results (rewards) through repeated trial and error. It is similar to a sales representative who tries various ways of proceeding with a proposal and applies the methods that received a good reaction from the other party to their next attempt. It is used in game strategy and controlling robot movements.

From Deep Learning to Generative AI

Now, let’s return to the inner box of the nesting structure.

Neural networks are a computational mechanism modeled after the connections of nerve cells in the human brain. They consist of a layer that receives input, layers that perform calculations in between, and a layer that outputs the answer. Deep learning is what happens when these intermediate layers are stacked deeply.Deep learning.

When layers are made deeper, the system becomes able to find what to focus on to make a judgment on its own, without humans having to teach it in detail. If we compare this to document screening, it is like a new employee who could only look at the checklist they were taught, but through experience, changes to be able to find the “points to look at” on their own.

And, based on the features of the data learned through deep learning, creating new text or images is the step forward to Generative AI. From AI that “distinguishes and predicts” to AI that “creates.” The flow from Chapter 1 to Chapter 2 can be understood through this single line.

Summary

AI, machine learning, and deep learning are nested in that order from the outside.
There are two ways to provide intelligence: “rule-based,” where humans write the rules, and “machine learning,” where the system learns from data.
Machine learning is divided into three types: supervised, unsupervised, and reinforcement learning.
Deep learning is a neural network with many layers stacked, and generative AI lies beyond that.

Course Information

From here on is information about the course material I have produced.

The content of Chapter 1 is covered in episodes 02 to 15 of the course. Starting with the definition of AI, we proceed in order through learning methods, neural networks, the AI boom, and the singularity, with each episode lasting around 7 minutes.

Generative AI Passport Exam Preparation Course (74 lectures in total, approx. 9 hours)

For a bridge to Chapter 2, please watch Episode 18 (VAE and GAN), which is available for free even before purchase. There are three other episodes available for free.

[Free] 10 Practice Questions | AI Basics and How Generative AI Works Review terms from Chapter 1 and Chapter 2 with 10 questions
Organizing the differences between GAN and VAE without using mathematical formulas Prevent confusion between generators and discriminators, and encoders and decoders
[First 100 people, free] Coupon for the video course A free coupon to take all 74 lectures (until 8:00 PM on October 27th, or while supplies last)

#GenerativeAIPassport #GenerativeAIPassportExam #GenerativeAIPassportExamPrep #AICertification #CertificationStudy #MachineLearning #DeepLearning

Source link